Application of neural network model based on similar day selecting in short-term load forecasting

Short-term load forecasting is the essential part of reliable and economic operation in power systems. In this paper, a new strategy suitable for selecting the training set for the neural network is presented. This strategy uses similarity degree parameters to identify the appropriate historical load data as training set for neural networks. This similar days selecting method can effectively avoid the problem of holiday and abrupt changes in influential factors, which make some historical load data unlikely for training the network. In addition, a neural network with a back propagation momentum training algorithm is proposed for load forecasting in order to reduce training time and improve convergence speed. The model validation is tested by Hebei province daily load data. Based on this model, the forecasting accuracy and learning potency are improved.